Blair · Key Terms
BlueForest Intelligence · plain-language definitions, illustrated
Blair is a single brain for BlueForest. Today the data is scattered across a dozen tools that can't see each other. Blair copies it all into one database, stamps everything with a standard Project ID so records about the same project, client or person can be matched up, and lets you ask questions in plain English across the whole picture. Everything hinges on one idea: a typed Project ID on everything.
The concepts everything else hangs on.
The whole system, and the future agent that will run it. Phase one is the deterministic system being built now: the Librarian, the store, a basic Analyst. Later comes Agentic Blair, an autonomous agent that maintains, reports and suggests upgrades.
Blair's collector and the deterministic heart of the system: automated, rules-based plumbing that gathers data from every source and keeps it organized and fresh. No autonomy; it does exactly what it's told.
Blair's question-answerer. Ask "what's the margin on the Carolina job?" and it answers from the unified data. Part deterministic (it writes the SQL for you), and the bridge to Agentic Blair's reasoning.
The counterpart to CLAUDE.md, but for Agentic Blair itself: a file defining Blair's motivations and priorities as an agent. What it should care about, watch for, and proactively propose.
XXX-YYYY-NNNThe single most important idea: a standard identifier stamped on every entity, like VID-2026-030. The 3 letters are the Division; the year and number make it unique. Letters first on purpose: sorts cleanly, survives spreadsheets. In the database this is the project_code field.
The master list at the heart of Blair: one row for every "thing" the business tracks. A video project, an internal system, a campaign, an employee. If it has a Project ID, it's in the registry, and everything else attaches to it. On the map, it's the interchange.
Any single item in the registry. Each has a Project ID, a name, a GitHub repo, a status, and a CLAUDE.md. A video project and an employee are both entities, just on different Division lines.
The category of an entity, shown by the 3-letter prefix of its Project ID. Four to start: VID video production, SYS internal systems & finance, MKT sales & marketing, EMP people.
The shared value that lets records from different systems be recognized as the same thing. Here the Project ID is the join key: a HubSpot deal, a Xero invoice and a NAS folder all carrying VID-2026-030 are understood to be one project.
The act of stitching those scattered records into one complete picture. The whole point of Blair: turning six half-views into one true view.
Why we're doing this.
One complete view of a client pulled from every system at once: deals, invoices, projects, emails, reviews. No more hunting through five tools.
The true profit and loss of a project: quoted (HubSpot) vs invoiced (Xero) vs labor cost (Streamtime time × rate). Blair can finally compute this because the code ties them together.
The top of the network: feeders, the express, and the moments data moves.
Any system where data is born (HubSpot, Xero, GitHub, BlueView…). The outer stations of the map.
A small automated program that logs into an external system and copies its data into Blair on a schedule. One per external source; together they're the Librarian's routes.
A BlueForest-built app that writes its data straight into Blair's database, so it needs no connector. These ride the tools express. They're the "Blue*" family below.
Self-hosted video review and hosting; replaces Vimeo. In build.
Video editing plus Remotion motion-graphics planning and execution. Planned.
Starts a new project: mints its Project ID, creates the repo and standard files, registers it. The on-ramp to the registry. Planned.
Marketing-content engine: video-first, spun into articles and podcast episodes. Planned.
A plain-text file in each project's repo describing that project: purpose, decisions, status. Blair ingests these so the Analyst knows the story of every project; a short version is auto-summarized for quick reference.
"Tell Blair the instant something changes." Used so CLAUDE.md updates land in real time.
"Check on a timer" (hourly, say). Used for routine API pulls.
A small program on the Mac Studio that catalogs the NAS and archive drives (which the cloud can't reach) and reports what's where. Shared by Blair and BlueView.
The data plumbing under the interchange.
The one database (Supabase / Postgres) where everything lands. One place, not twelve.
The "inbox": every piece of pulled data stored exactly as it arrived, never edited. A safety net; you can always rebuild from it.
The cleaned-up, organized versions (companies, projects, invoices…) derived from the raw data. What you actually query.
The blueprint of the database: what tables exist and how they connect.
A saved, ready-made query (like v_project_pnl) that shapes data usefully without storing a second copy.
Plain-language descriptions attached to every table and column, so an AI (or a new hire) knows what each field means. What makes the data "AI-readable."
row-level securityA database rule for who can see which rows. Used to keep EMP (employee) profiles owner-only.
How clean and trustworthy the data is. Example: HubSpot has ~22k contacts, many junk. Cleaning that is an early win.
How up-to-date the data is. Blair tracks when each source last synced and nudges you if something goes stale ("Streamtime budgets not updated in 3 weeks").
One line each. The stations you'll hear named.